Matching Images Using Invariant Level-line Primitives under Projective Transformation

This paper deals with a new registration method based on a specific level-line grouping. Because of its contrast-change invariance, our approach is an appropriate method for matching outdoor image sequences. Moreover, it does not require any estimation of the unknown transformation between images and handle well the critical cases that usually lead to pairing ambiguities, such as repetitive patterns in the images. This study focuses on invariants primitive construction under projective transformation, using level-lines. The registration by itself is performed through an efficient level-line cumulative matching based on a multi-stage primitive election procedure. Each stage provides a coarse estimate of the transformation that the next stage gets to refine. Experiments on real outdoor road scene show the accuracy and efficiency of this approach, using several image sequences covering different pertaining cases with different type of motion.